Abstract

With the advent of the incorporation of GPS receivers and then GPS-enabled smartphones in transportation data collection, many studies have looked at how to infer meaningful information from this data. Research in this field has concentrated on the use of heuristics and supervised machine learning methods to detect trip ends, trip itineraries, travel mode and trip purpose. Until now approaches to inference have relied on the use of fully-validated data. However, respondent burden associated with validation lowers participation rates and reduces the amount of precisely validated data because some people do not validate their trips or misreport them.
This thesis consists of two studies. In the first study I propose the application of a semi-supervised method to mode detection from smartphone travel survey data. Semi-supervised methods let researchers and planners use both validated and un-validated data. I compare the accuracy of three popular supervised methods (Decision Tree, Random Forest and Logistic Regression) with a simple semi-supervised method (Label Propagation with KNN kernel). Simple features such as speed, duration and length of trip and closeness of start and end points to transit network are used for model estimation. The results show that the semi-supervised method outperforms the supervised methods in the presence of high proportions of un-validated data and better predicts the observations in the test set. Furthermore, the run-time of the best model among the supervised methods was on average almost 16 times longer than the average run-times of the semi-supervised method.
In the second study, I develop a method to infer transit itineraries from smartphone travel survey data. Since the application of semi-supervised algorithms in travel surveys and transit itinerary detection are both in the early stages of development, a supervised approach is taken to tackle the problem of transit itinerary detection. To this end, trip features were extracted from smartphone collected data and transit network information available in the General Transit Feed Specification (GTFS) format. Based on these features, a Random Forest model was trained. Using the model, transit routes for 62% of trip segments was correctly detected.